Action Governance

Every governed decision is expressed in one canonical shape. An actor (a human, a service account, or an AI agent) attempts an action (pay an invoice, delete an IAM policy, export data) on a target (an invoice, a policy, a dataset) within a context (amount, signals, business facts). Fintra routes that request to the engine that owns the domain and returns a single verdict plus a trust score and a tamper-evident proof.

Different engines historically used different words - one said “deny recommended,” another “require approval.” Fintra normalizes them all onto these four, and any unrecognized verdict fails safe to hold, so a consequential action is never silently allowed.

No team lets a new system block production or hold a payment on the first day. Action Governance is built to be adopted in stages: it decides every action from the start, but what it does with the decision is a setting you raise deliberately.

1. 1Manual - a human decides; Fintra records the decision.
2. 2Shadow - Fintra decides and proves, but actuates nothing. You see exactly what it would have held.
3. 3Assisted - Fintra holds the risky items and releases the safe ones.
4. 4Enforced - Fintra acts on the verdict at the boundary.

A verdict you can’t prove later is a liability. Every decision is sealed with a canonical sha256 hash and written to the tenant-scoped Decision Ledger, so an auditor or an incident review can confirm who decided, on what basis, and that the record hasn’t been changed since.